(Here are my thoughts about playing this piece. Please add your thoughts in the comments at the bottom.)
I’m a little shy about writing about this piece, because it is perhaps Debussy’s most famous piano work. But I’ve been asked many questions about it in email, which inspired the following. Hopefully before long I’ll do a more complete writeup. In the meantime I hope the following helps.
Debussy does something very clever and subtle with the opening measures, both harmonically and in terms of the timing. To get the feel of the piece it helps to understand what Debussy is trying to do: Notice that the melody (based on the dominant d flat major chord) appears off of the downbeats through measure 8. In measure 9 for the first time we have the dominant chord on the downbeat, though it is really the middle of a phrase. The effect is very dreamy, with the melody being simultaneously very compelling but initially somewhat formless, taking more and more shape in the first 15 measures of the piece. After m. 15 things are more rhythmic though still rhythmically ambiguous, and we don’t have a really flowing melody until m. 27.
The effect Debussy is after is an initially dreamy melody that takes on more and more solidity. Lesser composers would have simply had poorly formed melodies in the beginning. What Debussy does is to have a fully formed, beautiful melody right away but he uses timing tricks to make it seem formless at first. These tricks are to have the melody appear off of the downbeats of the measures and to mix up triplet and eighth note rhythms so the rhythm of the melody is ambiguous. So you have asked the key question: how to you execute these tricks? I find that a very precise and strict counting in measures 1-14 provides exactly the effect that I think Debussy is aiming at. A more sloppy or improvised counting only mushes up the melody.
The technical trick is to be very comfortable mixing up triplets and eighth notes. (generally being able to play triplets against eighths is critical to playing Debussy’s music, but in this piece triplets and eighths appear at the same time only in m. 65.) A good way to practice this is to count triplets and eighths out loud, switching randomly between them. For example, “ONE two three ONE two three ONE and ONE and ONE two three ONE and ONE two three…” and so on with the “ONE”‘s on a regular beat. So the “ONE two three”‘s go faster than the “ONE and”s’. I find that I still have to slap my thigh on the “ONE”‘s to keep them on a steady beat. You might want to start by having a friend count out the “ONE”‘s so you don’t inadvertently count the “ONE and”‘s as two notes of the triplet. Then randomly mix up the triplets and the eighths, so you get used to switching back and forth while maintaining the steady beat on the “ONE”‘s. There are many such switches in the first 25 measures of Clair de Lune.
Once you are comfortable switching between triplets and eighths you can look at the music. The rule I use is: whenever the eighths are in a triplet (or for the dotted quarter/half notes) in the 9:8 time I count “ONE two three” and whenever the eighths are marked with a “2” they are played as “ONE and”. This is how I count them. I do not count up to nine. In all cases the ONEs fall on beats 1, 4, and 7 of each measure. So the first two measures are counted as three sets of “ONE two three” per measure. Yes, I do count the first eighth rest. Notice that in the first measure none of the notes falls on a “ONE”, leading to the dreamy effect. Then I count measure 3 as “ONE two three ONE and ONE and” of course not playing the tied note on the third “ONE”. Then we are in a straight “ONE two three” pattern until the bass in m. 8 then again with lots of two’s in m. 13 and 14. I count m. 13 and 14 just like m. 3. At measure 15 we enter a different pattern but built out of the same elements. It is to be played “rubato”, though, which means not quite as precisely as the previous measures. It is still good to count it out, and to put the rubato on a correct strict counting as very small variations in tempo. m. 15 is counted as “ONE and ONE two three ONE two three”, with no note on the second “ONE”, while m. 16 is “ONE two three ONE and ONE two three”. M. 17 and 18 are similar. M. 19, which is marked in my edition as one long sixth, I count as “ONE and ONE and ONE and”. Measures 19-24 are alternating between the “ONE and ONE and ONE and” and three sets of “ONE two three” per measure.
From that point on things are much more straightforward rhythmically, until the reprise at m. 51 which is almost a repeat of the openning. The only real surprise is m. 65, in which you have a “ONE two three” in the right hand and a “ONE and” in the left hand. In this case the “and” in the left hand should fall precisely between the “two” and “three” in the right hand. Before I had a good intuitive feel for this I would have counted it as “ONE two-and-three” which mixes the two modes of counting.
When listening to recordings, some pianists use the strict counting and some do not. I suppose you can follow your tastes, but everything I know about Debussy says that he was very strict about precise interpretation of his music. There is a nice story which goes something like: a pianist was visiting Debussy’s home and was playing a Debussy piano piece. At one point the pianist said “and this part is to be played freely, yes?”. Debussy later told a friend that he said nothing but looked down at the carpet and thought that this pianist would never tread on it again. So even when it seems that a section should be played freely, Debussy probably did not intend it that way unless he explicitly marked it otherwise. Besides, Debussy apparently went through a lot of trouble to create a rhythm which would sound free and improvised even when played in strict time. Seems to me if he went through that much trouble he meant it to be played that way.
Some other quick interpretational notes, without having looked recently at other people’s writings on this piece: I play the first 8 measure almost entirely without sustain pedal, holding the notes for their marked values, particularly the ties across the measure bars. Then I use pedal pretty liberally throughout the rest of the piece. I find the middle section very difficult, however, with respect to pedalling and voicing. I think it should be played very transparently, with each voice being very clear. I find this very difficult and more often than not I think I play it way too muddy. Because of this I think this is one of Debussy’s more difficult pieces: while other pieces have much more difficulty in hitting the notes, once I can hit them I play them fairly well. In the middle section of Clair de Lune I can hit the notes but I find I have to hit them EXACTLY right or the piece quickly deteriorates to mud.